As discussed in the book “Digital Watermarking”, by I. Cox et al., Morgan Kaufmann Publishers, Inc., San Francisco, 2001, information hiding or data hiding is a broad term related to making information imperceptible (as in a watermark), or keeping the information and its existence secret. Steganography, which comes from the Greek words meaning “to cover tightly” and “writing”, is the art and science of writing hidden or secret messages in such a way that no one, apart from the sender and intended recipient, suspects the existence of the message. For example, historical techniques to hide messages include the use of invisible inks or the writing of messages on envelopes in the area covered by postage stamps. As such, steganography is a form of security through obscurity. The presence of hidden data in a cover work maybe widely known, although not perceived, or the functional details can be known to just a few (people commonly know that currency has security features, although they know little of the specifics).
In the case of steganography, the hidden data is typically largely unrelated to the cover work in which it is hidden. By comparison, watermarking is the practice of altering a cover work to embed a message related to that work. The watermark can be a hidden message, such as embedded information in currency that aids authentication and thwarts counterfeiting. The watermark can also be obvious, such as data embedded in a cover work in a visually perceptible fashion. An example is a copyright protection message (e.g., text or logo) embedded in an image, with intent that the copyright holder gets attribution or payment for use of the image (cover work). In the modern world, steganography and watermarking are similar, and somewhat overlapping terms. Certainly, it is possible to embed both steganographic (unrelated) and watermark (related) hidden data in the same cover work.
Steganography and watermarking have both evolved technologically, including the emergence of digital steganography and digital watermarking. In digital steganography, data is concealed within a digital format, without causing noticeable changes to the file. Digital steganography is also related to hidden digital watermarking, although watermarking often uses smaller messages and has different purposes. In either case, a code that is imperceptible or nearly imperceptible to a human user can be embedded into media and detected by a machine vision system using an automated detection process. Typically, customized software is required to enable the practice of digital steganography or digital watermarking, in encoding, detecting, and decoding the hidden data.
It can be appreciated that there are many forms of cover work, including an image or text, a sound recording, an audio file, a document file, a video recording, or a software program. Although there are many known techniques for providing printed information with digital watermarks or steganographic data, opportunity remains for improved methods.
Printed digital watermarks, whether hidden or visible, are widely used to enable distinction of genuine items (e.g., media, currency, or artwork) from counterfeits and pirated copies. These authentication watermarks help combat losses of hundreds of billions of dollars in annual revenues that are stolen from industry by pirates (including counterfeiters). Alternate exemplary applications for hidden watermarks, whether embedded in an image or text, include providing a short sound track or an internet web address to an observer. Visible watermarks, such as those provided by commonly assigned U.S. Pat. No. 6,940,993 (Jones et al.), are often intended to be visually perceptible, although not visually objectionable, and can indicate the copyright owner of a given image. A QR code, which is a commonly used 2D bar code, can also be considered to be a form of a visible watermark.
The enablement of hidden digital watermarks is a trade-off of information capacity or density, image quality, and robustness. A digital watermark is considered fragile if it fails to be detectable after the slightest modification. Fragile watermarks are commonly used for tamper detection (integrity proof). Whereas, a digital watermark is semi-fragile if it is partially resistant to transformation, and a digital watermark is considered robust if it resists a designated class of transformations. Robust digital watermarks can be used for copy protection or to limit information access. Frequently, data for a given digital watermark is embedded in a multitude of images locations, to increase the likelihood that the hidden content can be reliably retrieved.
A variety of methods have been used to create hidden digital watermarks, including embedding hidden data within color images using frequency manipulation techniques that locally add a deliberate graininess to an image. Referring to FIG. 1, a typical prior art watermarking process 100 (or algorithm) is shown for embedding a message 105 into a digital image to create a watermarked work 140. In this case, the message 105 is the text “John Q. Public”. Since the message 105 will eventually be converted to a binary representation for processing in a computing device, the message 105 can be anything that can be represented in a binary code. For example, the message 105 can be another image, a sequence of map coordinates, an internet address, etc. The cover work 110 is the digital image in which the message 105 will be embedded. The cover work 110 can be specific to an application or chosen at random, specifically to enhance some aspect of watermarking performance, or for any number of reasons unrelated to the watermarking application itself.
In many watermarking systems security is important, and an encryption algorithm creates an encrypted message 115 from the message 105 as a first step in the watermarking process 100. In FIG. 1, this is illustrated by a simple cipher in which each letter in the original message is substituted by the next letter in the alphabet. It should be obvious to a practitioner in the field that any cipher can be used as long as recipients decrypting the watermarked work 140 have knowledge of the cipher. After the message is encrypted, an encoding step follows to encode the encrypted pixel data to provide an encoded message 120. This encoded pixel data can be an image that is the same size as the cover work 110 or a smaller image (i.e. tile) that is some fraction of the size of the cover work 110.
The amplitude of the encoded message is then reduced to conform to the goals of the practitioner(s). The resulting reduced amplitude encoded message 125 is then added to the cover work 110 to produce a watermarked work 140 (a cover work 110 with embedded message 130). If the reduced amplitude encoded image 125 is a tile, the addition operation can involve adding the tile to the image at a multiplicity of non-overlapping locations in the cover work 110. In many watermarking systems, attributes of the cover work 110 are used in the encoding and amplitude reduction steps of the watermarking algorithm. The resulting watermarked work 140 can then be printed or displayed.
In a typical watermarking system, practitioners have a number of goals. Four important goals are: low visibility of the embedded message for individuals that are not intended recipients of the message; high perceived image quality of the cover work 110 once the message 105 has been transformed into an embedded message 130; robustness of that embedded message 130 to subsequent changes in the cover work 110 carrying with embedded message 130; and high information content of the embedded message 130. The importance of these goals to the practitioner in a particular watermarking algorithm affect the parameters used to control the encrypting, encoding, and amplitude reduction steps of the watermarking process 100. Often practitioners must trade off the goals of reduced detectability and high image quality against the competing goals of robustness of the embedded message and high information content in the embedded message. It is often desirable to reduce the amplitude of the encoded message 120 in order to make the watermark less detectable and to increase the perceived image quality of the cover work 110 with the embedded message 130.
For example, commonly-assigned U.S. Pat. No. 5,859,920 (Daly et al.) provides a watermarking approach where data is embedded in the source image data using a spatial frequency dispersal method. This process involves convolving the image data with an encoding carrier image to produce a frequency dispersed data image. As another example, U.S. Pat. No. 6,683,966 (Tian et al.), describes a watermarking method in which media signal is transformed from its perceptual domain to frequency domain regions and watermark data is embedded into one or more frequency domain regions. The image is restored to a perceptual space, in which it can be printed or displayed with the hidden data. Alternately, U.S. Pat. No. 6,522,767 (Moskowitz) provides optimization methods for the insertion, protection, and detection of digital watermarks in digitized data. In particular, the quality of the underlying content signals can be used to identify and highlight advantageous locations for the insertion of digital watermarks. The watermark is integrated as closely as possible to the content signal, at a maximum level to force degradation of the content signal when attempts are made to remove the watermarks. For example, this can mean locating intensity changes that represent the watermark, and which can appear as noise to a human viewer, in image locations that have variable content instead of locations that have uniform or nearly uniform content.
As another example, where color coding is used instead of frequency coding, U.S. Pat. No. 8,064,100 (Braun et al.) provides a method for hiding a watermark in a color image using colorants that have essentially the same visual color as other colorants, but which have different spectral profiles in the visible spectrum. The difference between these metameric pair colorant sets can be revealed by illuminating the color image with narrow bandwidth visible light sources such as light emitting diodes (LEDs).
A problem with embedding watermarks in images to be sufficiently imperceptible to human viewers, such that they are reliably hidden, is that the information capacity and density are often quite limited. For example, a typical digital watermark has a constrained data capacity of only 32 bits, which is generally insufficient to directly store a website link. Instead, such a watermark can provide a smaller data code that can be translated at a database, such as at the Discover Online Services Portal™, provided by Digimarc (Beaverton, Oreg.). By comparison, the highly visible QR codes can store between 100 bits to 20 kbits of data, depending on the visible area occupied by the code.
Alternately, digital watermarks or covert data can be provided by printing invisible or nearly invisible features. For example, AlpVision (Vevey, SW) offers a solution called Cryptoglyph™, in which micro-dots are printed with standard inks with the dot size and dot color manipulate to make them virtually invisible. For example, a small yellow dot can be hard to perceive on white paper, being visually lost within the imperfections of the paper, and yet also be detected by a flat-bed document scanner. While this solution does not require special inks, the hidden data density is low and the data cannot be embedded in busy images.
Digital data can also be hidden using “invisible” inks, using materials with at least an absorbance outside the visible band. However, such inks typically have visible color crosstalk and can only be used under limited conditions. For example, the paper “Invisible Marker Based Augmented Reality System”, by H. Park and JA. Park, published in the SPIE Proc., Vol. 5960 (2005), used an infrared (IR) ink that absorbs 793 nm IR light and fluorescently emits at 840 nm IR light. However, as the ink had a faint green appearance, ink density needs to be low to retain invisibility. Additionally, both a specialized light source and imaging device are required. As another example, the paper “Formulation of an Invisible Infrared Printing Ink”, by M. Yousaf & M. Lazzouni, published in Dyes and Pigments, Vol. 27, pp. 297-303 (1995), discusses the use of a silicon naphthalocyanine based IR absorbing ink, which provides small light absorption change (ΔR˜15%) at 790 nm, that can be detected using an illumination source emitting at 790 nm. This small IR absorption is limited in part by visible crosstalk, as the dye is not truly invisible, but has a green tint. Additionally, crosstalk of visible dye absorption into the IR spectrum could easily mask or confuse this weak IR absorption signal. Increasing the IR absorption density to improve signal detection or bit depth is limited in part because the crosstalk visible absorption increases as well. Yousaf et al. suggests that the green tint can be overcome by printing the “invisible” infrared printing ink on a uniform green tinted print media.
It is also noted that IR dyes or pigments are particularly unstable, typically when used in low concentrations, and vulnerable to environmental degradation, including high humidity or dye fade with UV or visible light exposure. Commonly-assigned U.S. Pat. No. 6,706,460 (Williams et al.) describes an ameliorative process that involves loading IR dye in a latex particle.
Invisible ultraviolet inks have also been used for stegonagraphy and watermarking. In particular, inks in which incident ultraviolet light (UV) stimulates visible fluorescence can be particularly useful. For example, U.S. Pat. No. 5,542,971 (Auslander et al.) describes an ultraviolet ink composition that provides visible fluorescence in response to UV exposure that can be used to print bar code information. Typically, such inks are revealed by illumination from specialized light sources such as black lights or UV LEDs that provide UVA (315-400 nm) or UVB (280-315 nm) light. However, as atmospheric filtered solar UV light extends down to 280 nm, there is a risk for accidental activation and content disclosure. As the solar intensity in the UVB range is comparatively low, the risk of accidental disclosure is reduced for materials with lower activation wavelengths.
Of course fluorescent materials that both absorb and emit not visible light, whether UV or IR (see U.S. Pat. No. 6,149,719 (Houle) for example), can used to reduce visible spectrum visibility, but then both a special illuminant and imaging device are required. As another approach to compensating for the visible fluorescence of UV stimulable materials, U.S. Pat. No. 6,718,046 (Reed et al.) provides a low visibility watermark using time decay fluorescence. In particular, this fragile digital watermark can be printed with two UV inks having visible fluorescence with different (short and long) decay times. When stimulated with a UV pulse, the digital watermark is detected after the first emission decay time, but before the second emission decay time. However, when illuminated by a constant or steady-state UV illumination, such as atmospheric filtered solar UV light, both materials fluoresce, obscuring the watermark. Although such approaches can be desirable for very covert applications, they are less desirable for consumer applications.
In summary, opportunity remains to provide improved solutions for steganography or watermarking that enable the embedding of comparatively large amounts of hidden data in images, while causing minimal perceptible degradation in image quality for human observers. Additionally, embedded watermarks of this type, whether digital or analog, that can be easily detected and interpreted by consumer devices, would have significant value.